Partial Rule Weighting Using Single-Layer Perceptron

Inductive Logic Programming (ILP) has been widely used in Knowledge Discovery in Databases (KDD). The ordinary ILP systems work in two-class domains, not in multi-class domains. We have proposed the method which is be able to help ILP in multi-class domains by using the partial rules extracted from the ILP’s rules combined with weighting algorithm to classify unseen examples. In this paper, we improve the weighting algorithm by using single layer perceptron. The learned weights from the perceptrons and the partial rules are then combined to represent the knowledge extracted from the domains. The accuracy of the proposed method on classification of a real-world data set, dopamine antagonist molecules, shows that our approach can remarkably improve the previous weighting algorithm and the original ILP’s rules.